Unified Multimodal Model for Brain MRI Imputation and Understanding

The article presents UniBrain, a unified multimodal model designed for brain MRI analysis, which addresses the challenges of missing data in medical imaging. It employs a novel training strategy that combines imaging modality imputation and understanding through an autoregressive approach, enhanced by a self-alignment strategy for learning anatomical features without detailed captions. UniBrain demonstrates high performance in brain image imputation and disease diagnosis across diverse levels of modality incompleteness, making it a significant advancement for practitioners working with multimodal medical data.